Commit History

  1. Add few-shot prompt and shaped rewards

    c166ffe
  2. Revert to Gradio 5.49.1 and ASCII logs

    c5c47e3
  3. Pin datasets for hub compatibility

    75d63f1
  4. Quote python_version

    bf7b2ba
  5. Use Gradio 4.44.1 on Python 3.10

    96da636
  6. Pin huggingface_hub for gradio4

    55aa775
  7. Use audioop-lts for py3.13

    c8f2b1f
  8. Add pyaudioop for Python 3.13

    4294719
  9. Pin Gradio to 4.x

    8796c0b
  10. Disable Gradio queue for buttons

    13b4a50
  11. Replace timer polling with manual Refresh Log

    4da69c3
  12. Disable queue for UI polling and buttons

    4daeb52
  13. Initialize log immediately on Start

    b8e25bb
  14. Fix resume adapter training (no peft_config with PeftModel)

    7aa68cd
  15. Fix code block extraction and increase completion length

    1ee2461
  16. Persist LoRA adapter via HF dataset repo

    f2d5eaa
  17. Switch to correctness-gated GRPO LoRA with persistence

    648e193
  18. Add VLIW simulator for cycle-count based rewards

    b3b926b

  1. Add GRPO training with proper state management

    9c10799
  2. Simpler test version for model loading

    a7473b9
  3. Add training functionality with GRPO

    cccd413
  4. Minimal test version

    e721a4b
  5. Simplified version without simulator

    b03b587
  6. Full training app with verified imports

    77e3392
  7. Debug version to check imports

    f4dcd8f
  8. Fix TRL and dependencies versions

    61f3fc6
  9. Initial commit: VLIW kernel optimizer via RL

    1d07708
  10. initial commit

    Verified signature
    3aa84d6